Abstract
Well constructed chronologies are critical to reliable paleoecological inference. The systematic construction of age models from the Neotoma Paleoecological Database provides the opportunity to examine the assumptions and decisions involved in model construction. We use the Bacon age modeling software, implemented in R and attempt to catalogue the decision-making process neccessary for chronology development. This paper addresses 210Pb uncertainty, the treatment of inflection points in sedimentation rates, the use of Ambrosia as a biostratigraphic marker in North America, settings for accumulation rates, memory parameters, section and thicknesses within Bacon, and ultimately compare Bacon outputs to outputs from theBChron R package. Researchers are increasingly using large paleoecological databases, either in their entirety, or as localized data subsets, to either undertake synthetic analysis, or to add breadth to their new findings (Brewer et al., 2012; Uhen et al., 2013). As many paleoecologists are aware however, the chronology assigned to a sedimentary archive can strongly affect the interpretation of a record. Paleoecologists have been aware of limitations in age modeling capabilities for some time, and the issues that these limitations cause are well noted (Grimm et al., 2009; Liu et al., 2012).
Recent efforts have focused on standardizing age models across the database (Giesecke et al., 2014), establishing regional benchmarks (Blois et al., 2011; Flantua et al., 2016). European models used the software Clam (Blaauw, 2010), as did efforts with the North American Pollen Database (Blois et al., 2011).
As more and more ecologists are turning to the paleoecological record, and as we try to do more with what we have, the limitations of extant age models become more problematic. Indeed, it is possible to wiggle match many different patterns, and the flexibility in modeling age-depth relationships can introduce additional “researcher degrees of freedom” that might lead to greater rates of false-positive relationships in paleoecological research (Blaauw, 2012).
The Neotoma database (Grimm, 2008) contains 1864 global pollen records that can be used for paleoecological analysis. These records have been obtained from publications that span a time period from 1948 to 2016, with more than half the records coming from before 1983. For pollen records that are not “modern”, sample age is obtained from a chronology constructed using classical (Blaauw, 2010) or Bayesian (Blaauw and Christen, 2011, 2005; Buck and Sahu, 2000; Ramsey, 1995) methods using dated material including radiocarbon (14C) and other radiometric dates (e.g. 210Pb, 137Cs).
Chronologies are developed using dated stratigraphic control points from within a cores, which may be geochronological (dated material), geostratigraphic (e.g., the “modern” core top), and sometimes biostratigraphic (changes in pollen assemblages associated with dated changes on the landscape). Geochronological control point ages are often uncertain due to analytical errors during the laboratory radiocarbon dating process (Ward and Wilson, 1978), the conversion of radiocarbon to calendar years (Reimer et al., 2013), and potential differences between the ages of macrofossil material and age of sediment (Blois et al., 2011). Geostratigrahic markers may have fewer sources of uncertainty – the core top age is assumed to be the year of sampling – although sedimentary mixing of the upper sediment during sampling does introduce some uncertainty. Finally, biostratigraphic control points are determined by the examination (usually visual) of changes in pollen assemblages throughout a core. However, time series of pollen counts are noisy, and in practice identifying changes in composition is both difficult and subjective (Dawson et al., 2016). There remains uncertainty in the identification of compositional shifts related to the chronological/stratigraphic sampling density (Liu et al., 2012), and, in the case of landscape-scale phenomena such as the mid-Holocene Hemlock decline (Bennett and Fuller, 2002; Davis, 1981), the need to assign temporal bounds to the landscape-scale phenomenon that caused the compositional shift.
The development of the first IntCal curve was a major milestone in paleoecological analysis (Hughen et al., 1998). IntCal98, and subsequent itterations, allowed researchers to move from radiocarbon years to calibrated radiocarbon years using material of known age and associated 14C dates to build a calibration curve. Because radiocarbon years are not equivalent to calendar years, and because the relationship is non-linear, the use of calibration curves have provided researchers with an important tool to help improve model chronologies. However, across the Neotoma database (here we refer to North America only), many records within the database still record chronologies using only radiocarbon years (Figure 1).
The transition from age models using only radiocarbon years to those with calibrated radiocarbon years within Neotoma is dramatic. The final radiocarbon model appears to be from 1998, following this we see no more radiocarbon models. Along with this transition, we are seeing a second transition from simple linear models to more complex models using flexible Bayesian methods. A critical question then becomes, when faced with records generated using only uncalibrated dates, should we calibrate radiocarbon dates, generate age models de novo, or ignore the records altogether?
Figure 1. Number and type of default chronologies for pollen records in the Neotoma Paleoecological Database based on the original year of publication for the dataset. Within Neotoma the default dataset chronology may be updated by subsequent researchers when this information is provided to a data steward.
To understand the importance of developing new chronologies, it is important to understand that there can be a significant difference in interpreted ages when either, chronological control points are re-calibrated and a model is generated using those calibrated controls, or, the interpolated ages from a model are directly re-calibrated from radiocarbon years to calendar years before present. While not the prefered method, direct recalibration of interpolated ages does occur within the Neotoma ecosystem. For example, the temporal search function within the Neotoma Explorer (http://apps.neotomadb.org/explorer), Tilia (http://tiliait.org) and the Neotoma API (http://api.neotomadb.org) all use a lookup table that directly recalibrates ages in radiocarbon years. However, this process results in systematic biases in both synthetic data (Figure 2) and in the actual Neotoma data (Figure 3). flexible]
Figure 2. The difference between directly calibrating radiocarbon ages from an existing model and rebuilding age models using linear interpolatation from calibrated radiocarbon ages. Negative values result when the rebuilt model provides older dates than the directly calibrated dates. Negative values indicate that recalibration of interpolated 14C dates followed by linear interpolation of calibrated ages provides systematically older ages than direct recalibration of interpolated ages obtained from a dataset with linearly interpolated dates reported in radiocarbon years.